HoloTalk by Holodata: the film
One speaker at the lectern, an audience reading along in Korean, Simplified Chinese, Vietnamese and other languages on their own phones, and a summary in each reader's language afterwards. The film has no spoken words: everything it says is on screen, and its captions give its words and its sounds.
On this page
The film
The film's captions are in English; turn them on in the player.
In the film
- 0:00One speaker, a thousand listeners
- 0:13Every word, every language, as it's spoken
- 0:29Built for real rooms
- 0:43After the session
The figures in the film and their conditions
In the film: 0.67 s is the median from pause to final text in HoloTalk's speech benchmark (October 2026); 99.6% is character accuracy over an hour of synthetic Mandarin speech with clean audio; 4.7+ out of 5 means every translation model HoloTalk uses averaged 4.7 or more over 55 test sentences, scored by an AI judge (HoloTalk's translation eval, October 2026).
What happens in the film
HoloTalk, a 60-second film. In a dark hall, a point of light becomes the HoloTalk mark: one speaker, a thousand listeners. Three of them, Mina, Wei and Linh, cannot follow the speaker's English, and as "Welcome to the Sydney Innovation Summit." reaches them, the words fall apart. Every word, every language, as it's spoken. The words appear as they are spoken, and the final text lands 0.67 seconds after each pause (the median, from pause to final text, in HoloTalk's speech benchmark, October 2026). The sentence's light splits into ten languages at once (HoloTalk has 45 caption languages, 10 at once) and reaches their phones: Mina reads it in Korean, Wei in Chinese and Linh in Vietnamese, on their own phones with no app, and the whole hall lights up. In real rooms, names are spelled right, one sentence in three languages is captioned, and when the network drops nothing is lost (up to 5 minutes of audio is kept on the speaker's device and sent on reconnect; longer outages are added after the meeting). Heard right: 99.6% character accuracy over an hour of synthetic Mandarin speech with clean audio. Translated right: every translation model HoloTalk uses averaged 4.7 or more out of 5 over 55 test sentences, scored by an AI judge (HoloTalk's translation eval, October 2026). After the meeting, the summary reaches each of them in their own language, in Seoul, Shanghai and Ho Chi Minh City, and Mina asks, in Korean, what was decided: the answer comes in Korean, citing the words spoken. HoloTalk: every word, every language, as it's spoken. holotalk.au. Powered by Holodata.
About the film
The people and the event in the film are fictional. The captions and translations they read are HoloTalk's own output; the answer to Mina's question was written for the film and translated by HoloTalk.
Captions and translations are generated automatically and can contain errors; where accuracy is critical, such as medical or legal content, also book a qualified interpreter.
How it works explains a session step by step, and the FAQ answers the usual questions.